Re: Sequence being released?

[Michael Meeks <michael ximian com>]

Hi Chris,

On Sun, 2003-09-14 at 12:11, Chris Roberts wrote:
> However, on the second call to this method, the _buffer variable is NULL.
> I've tried different ways of allocating the sequence, using the CORBA calls,
> and malloc and such, but it seems to me that ORBit is ignoring the _release
> flag and freeing the buffer anyway.

	That would suck; the code that implements this is in:

	ORBit2/src/orb/orb-core/allocators.c (ORBit_freekids_via_typecode) -
and it looks alright to me:

	case CORBA_tk_sequence: {
		CORBA_sequence_CORBA_octet *pval = mem;
		if (pval->_release)
			ORBit_free_T (pval->_buffer);
		pval->_buffer = NULL;
		retval = (guchar *)mem + sizeof(*pval);
		break;
	}

	It paranoidly NULL's the copy you made's ->_buffer - but that shouldn't
affect the original at all.

	Of course - we have to free the toplevel 'sequence' itself; I _believe_
release is only there to avoid re-allocating the _buffer rather than the
whole sequence; thus you need:

	aRetval = foo_alloc;
	aRetval->_release = FALSE;
	aRetval->_buffer = main_foo->_buffer;
	aRetval->...

	Of course - unless this buffer is _Really_ static it's an incredibly
risky thing to do to poke with this, it's fraught with problems.

	Particularly if a re-entering call updates / reallocs / moves _buffer
while the ORB is still trying to marshal the return value [ lower down
the stack perhaps ], or a similar situation with another thread - makes
it almost always better/cleaner just to dup/return.

	HTH,

		Michael.

-- 
 michael@ximian.com  <><, Pseudo Engineer, itinerant idiot